Qwen Code is a direct fork of Gemini CLI, and that fact explains almost everything else about this environment: its behavior, its file-naming conventions and its general delegation logic closely mirror what was already described in the Gemini section. Project memory is set by a QWEN.md file - the name is not hardcoded but a convention configured through context.fileName, though by default it plays the same role GEMINI.md plays in the original environment. Settings live in .qwen/settings.json, and the orchestration itself is assembled from a combination of tools and subagents rather than from a separate SDK. The model behind the environment is Qwen3-Coder.
In practice this means that nearly all the techniques from the Gemini section carry over here almost verbatim: the same project memory-file principle, the same notions of what a tool is and what a subagent is, the same overall pipeline of reading code, planning, building and review. The difference is concentrated in two places - the model running under the hood, and the fact that some capabilities and configuration details depend on the specific fork version rather than being a general property of Qwen Code as a product.
That is exactly why this section does not tie specific techniques to a specific Qwen Code version number: the fork evolves at its own pace relative to Gemini CLI, and what holds true today is worth explicitly checking against current documentation rather than relying on memory of an earlier version.
The official resources here are the Qwen Code repository on GitHub, its documentation, a separate architecture overview, and the Qwen3-Coder model card; together they cover both the practical and the architectural side of the environment.
The meta-prompt sets QWEN.md to the orchestrator role with its own stop conditions, describes the available tools and subagents, and runs the same pipeline - explore, plan, build, review - as in the Gemini section, while explicitly noting that Qwen Code is a Gemini CLI fork and so its behavior is worth checking against the original wherever Qwen Code's own documentation gives no direct answer. The plan, as everywhere, is shown before changes are made.
Read the Qwen Code docs:
- qwenlm.github.io/qwen-code-docs
- github.com/QwenLM/qwen-code
Set up THIS project:
1. Create/update QWEN.md (name by convention, see context.fileName;
project memory, like GEMINI.md):
role - orchestrator; delegation rules; stop conditions.
2. Describe tools and subagents; pipeline
explore -> plan -> build -> review.
3. Balance quality vs token economy: narrow context,
summaries instead of dumps, no extra passes.
Qwen Code is a Gemini CLI fork, behaviour is similar.
Show the plan BEFORE changes.